3 ms·Universal and Transferable Adversarial Attacks on Aligned Language Models14 points by llazzaro 3y ago